1) The study examined the effects of probiotic supplementation with Lactobacillus paracasei or Lactobacillus rhamnosus on gut bacteria composition and host metabolism in humanized extended genome mice. 2) Probiotic exposure altered the gut microbiome and resulted in changes to hepatic lipid metabolism, lowered plasma lipoprotein levels, and stimulated glycolysis. It also impacted amino acid metabolism, methylamines, and short-chain fatty acid levels. 3) Multivariate analysis of metabolic profiles from multiple compartments, including biofluids, tissues, and cecal contents, provided a systems-level view of the host response to probiotic interventions. This integrated approach demonstrated how probiotics can

1. Molecular Systems Biology 4; Article number 157; doi:10.1038/msb4100190
Citation: Molecular Systems Biology 4:157
& 2008 EMBO and Nature Publishing Group All rights reserved 1744-4292/08
www.molecularsystemsbiology.com
Probiotic modulation of symbiotic gut microbial–host
metabolic interactions in a humanized microbiome
mouse model
Francois-Pierre J Martin1,2, Yulan Wang1, Norbert Sprenger2, Ivan KS Yap1, Torbjorn Lundstedt3,4, Per Lek3, Serge Rezzi2, Ziad
¨
Ramadan2, Peter van Bladeren2, Laurent B Fay2, Sunil Kochhar2, John C Lindon1, Elaine Holmes1 and Jeremy K Nicholson1,*
1
Department of Biomolecular Medicine, Division of Surgery, Oncology, Reproductive Biology and Anaesthetics, Faculty of Medicine, Imperial College London, London,
UK, 2 Nestle Research Center, Lausanne, Switzerland, 3 AcurePharmaAB, Uppsala, Sweden and 4Department of Medicinal Chemistry, BMC, Uppsala University,
´
Uppsala, Sweden.
* Corresponding author. Department of Biomolecular Medicine, Division of Surgery, Oncology, Reproductive Biology and Anaesthetics, Faculty of Medicine, Imperial
College London, Sir Alexander Fleming Building, South Kensington Campus, London SW7 2AZ, UK.
Tel.: þ 44 20 7594 3195; Fax: þ 44 20 7594 3226;
E-mail: j.nicholson@imperial.ac.uk
Received 29.6.07; accepted 17.10.07
The transgenomic metabolic effects of exposure to either Lactobacillus paracasei or Lactobacillus
rhamnosus probiotics have been measured and mapped in humanized extended genome mice
(germ-free mice colonized with human baby flora). Statistical analysis of the compartmental
fluctuations in diverse metabolic compartments, including biofluids, tissue and cecal short-chain
fatty acids (SCFAs) in relation to microbial population modulation generated a novel top-down
systems biology view of the host response to probiotic intervention. Probiotic exposure exerted
microbiome modification and resulted in altered hepatic lipid metabolism coupled with lowered
plasma lipoprotein levels and apparent stimulated glycolysis. Probiotic treatments also altered a
diverse range of pathways outcomes, including amino-acid metabolism, methylamines and SCFAs.
The novel application of hierarchical-principal component analysis allowed visualization of
multicompartmental transgenomic metabolic interactions that could also be resolved at the
compartment and pathway level. These integrated system investigations demonstrate the potential
of metabolic profiling as a top-down systems biology driver for investigating the mechanistic basis of
probiotic action and the therapeutic surveillance of the gut microbial activity related to dietary
supplementation of probiotics.
Molecular Systems Biology 15 January 2008; doi:10.1038/msb4100190
Subject Categories: metabolic and regulatory networks; microbiology and pathogens
Keywords: metabonomics; microbiome; NMR spectroscopy; probiotics; UPLC-MS
This is an open-access article distributed under the terms of the Creative Commons Attribution Licence,
which permits distribution and reproduction in any medium, provided the original author and source are
credited. Creation of derivative works is permitted but the resulting work may be distributed only under the
same or similar licence to this one. This licence does not permit commercial exploitation without specific
permission.
Introduction 2005; Gill et al, 2006; Ley et al, 2006). Mammalian–microbial
symbiosis can play a strong role in the metabolism of
The gut microbiome–mammalian ‘Superorganism’ (Leder- endogenous and exogenous compounds and can also be
berg, 2000) represents a level of biological evolutionary influential in the etiology and development of several diseases,
development in which there is extensive ‘transgenomic’ for example insulin resistance (Dumas et al, 2006), Crohn’s
modulation of metabolism and physiology that is a character- disease (Gupta et al, 2000; Marchesi et al, 2007), irritable
istic of true symbiosis. By definition, superorganisms contain bowel syndrome (Sartor, 2004; Martin et al, 2006), food
multiple cell types, and the coevolved interacting genomes can allergies (Bjorksten et al, 2001), gastritis and peptic ulcers
only be effectively studied as an in vivo unit in situ using top- (Warren, 2000; Marshall, 2003), obesity (Ley et al, 2006;
down systems biology approaches (Nicholson, 2006; Martin Turnbaugh et al, 2006), cardiovascular disease (Pereira and
et al, 2007a). Interest in the impact of gut microbial activity on Gibson, 2002) and gastrointestinal cancers (Dunne, 2001).
human health is expanding rapidly and many mammalian– Activities of the diverse gut microbiota can be highly specific
microbial associations, both positive and negative, have been and it has been reported that the establishment of Bifidobacteria
reported (Dunne, 2001; Verdu et al, 2004; Nicholson et al, is important for the development of the immune system and
& 2008 EMBO and Nature Publishing Group Molecular Systems Biology 2008 1

2. Probiotics modulation of mammalian metabolism
F-PJ Martin et al
for maintaining gut function (Blum and Schiffrin, 2003; mathematical modelling is a tool which is well suited to
Salminen et al, 2005; Ouwehand, 2007). In particular, elevated generate metabolic profiles that encapsulate the top-down
counts in Bifidobacterium with reduced Escherichia coli, system response of an organism to a stressor or intervention
streptococci, Bacteroides and clostridia counts in breast-fed (Nicholson and Wilson, 2003). Multivariate metabolic profil-
babies compared to formula-fed neonates may result in the ing offers a practical approach to measuring the metabolic
lower incidence of infections, morbidity and mortality in endpoints that link directly to whole system activity and which
breast-fed infants (Dai et al, 2000; Kunz et al, 2000). As the are determined by both host genetic and environmental factors
microbiome interacts strongly with the host to determine the (Nicholson et al, 2005). Recently, metabolic profiling strategies
metabolic phenotype (Holmes and Nicholson, 2005; Gavaghan have been successfully applied to characterizing the metabolic
McKee et al, 2006) and metabolic phenotype influences consequences of nutritional intervention (Rezzi et al, 2007;
outcomes of drug interventions (Nicholson et al, 2004; Clayton Wang et al, 2007) the effects of the gut microflora on
et al, 2006), there is clearly an important role of understanding mammalian metabolism (Martin et al, 2006, 2007a, b) and
these interactions as part of personalized healthcare solutions mechanisms of insulin-resistance (Dumas et al, 2006). In the
(Nicholson, 2006). current study, 1H nuclear magnetic resonance (NMR) spectro-
One of the current approaches used to modulate the balance scopy and targeted ultra performance liquid chromatography-
of intestinal microflora is based on oral administration of mass spectrometry (UPLC-MS) analysis have been applied to
probiotics. A probiotic is generally defined as a ‘live microbial characterize the global metabolic responses of humanized
feed supplement which beneficially affects the host animal by microbiome mice subsequently exposed to placebo, Lactoba-
improving its intestinal microbial balance’ (Fuller, 2004). The cillus paracasei or Lactobacillus rhamnosus supplementation.
gastrointestinal system is populated by potentially pathogenic Correlation of the response across multiple biofluids and
bacteria that are capable of degrading proteins (putrefaction), tissue, using plasma, urine, fecal extracts, liver tissues and
releasing ammonia, amines and indoles, which in high ileal flushes as the biological matrices for the detection of
concentrations can be toxic to humans (Cummings and dietary intervention, generates a top-down systems biology
Bingham, 1987). Probiotic supplementation aims at replacing view of the response to probiotics intervention.
or reducing the number of potentially harmful E. coli and
Clostridia in the intestine by enriching the populations of gut
microbiota that ferment carbohydrates and that have little Results
proteolytic activity. Probiotics, most commonly Lactobacillus
Gut bacterial composition
and Bifidobacteria, can be used to modulate the balance of the
intestinal microflora in a beneficial way (Collins and Gibson, Microbiological analyses were performed on fecal samples to
1999). Although Lactobacilli do not predominate among the assess the growth of the HBF in germ-free mice and to ascertain
intestinal microflora, their resistance to acid conditions and the effects of probiotics on the development of gut bacteria.
bile salts toxicity results in their ubiquitous presence through- The measured terminal composition of the fecal microbiota is
out the gut (Corcoran et al, 2005), hence they can exert detailed in Table I, where the statistically significant differences
metabolic effects at many levels. Fermented dairy products between the various groups were calculated using a two-
containing Lactobacillus have traditionally been used to tailed Mann–Whitney test. The bacterial populations of
modulate the microbial ecology (Dunne, 2001). In particular, Bifidobacteria longum and Staphylococcus aureus were reduced
L. paracasei was shown to modulate the intestinal physiology, after introduction of both probiotics. Additionally, unique
to prevent infection of pathogenic bacteria (Sarker et al, 2005), effects of L. rhamnosus supplementation caused decreased
to stimulate the immune system (Ibnou-Zekri et al, 2003), and populations of Bifidobacterium breve, Staphylococcus epidermi-
to normalize gastrointestinal disorders (Martin et al, 2006). dis and Clostridium perfringens but an increase of E. coli.
L. rhamnosus is also a significant probiotic strain with proven
health benefits and therapeutic applications in the treatment of
Gut levels of short-chain fatty acids
diarrhea (Szynanski et al, 2006), irritable bowel syndrome
(Kajander et al, 2005), atopic eczema (Corcoran et al, 2005) Short-chain fatty acids (SCFAs), namely acetate, propionate,
and the prevention of urinary tract infections (Reid and Bruce, isobutyrate, n-butyrate and isovalerate, were identified and
2006). However, the functional effects of probiotic interven- quantified from the cecal content using GC-FID. The results,
tions cannot be fully assessed without probing the biochem- presented in Table II, are given in mmol per gram of dry fecal
istry of the host at multiple compartmental levels, and we material and as mean±s.d. for each group of mice. The
propose that top-down systems biology provides an ideal production of some of the SCFAs, that is, acetate and butyrate,
approach to further understanding in this field. The microbiota by the HBF mice supplemented with both of the probiotics
observed in human baby flora (HBF) mice have a number of was reduced. In addition, increases of the concentrations in
similarities with that found in formula-fed neonates (Mackie isobutyrate and isovalerate were observed in the mice fed with
et al, 1999), which makes it to be a well-adapted and simplified L. paracasei.
model to assess probiotics impact on gut microbial functional
ecosystems (in particular on metabolism of Bifidobacteria
and potential pathogens) and subsequent effects on host
Analysis of1H NMR spectroscopic data on plasma,
metabolism. urine, liver and fecal extracts
Metabolic profiling using high-density data generating A series of pairwise O-PLS-DA models of 1H NMR spectra were
spectroscopic techniques, in combination with multivariate performed to extract information on the metabolic effects of
2 Molecular Systems Biology 2008 & 2008 EMBO and Nature Publishing Group

3. Probiotics modulation of mammalian metabolism
F-PJ Martin et al
Table I Microbial species counts in mouse feces at the end of the experiment
Groups/log10 CFU HBF (n=10) HBF+L. paracasei (n=9) HBF+L. rhamnosus (n=9)
L. paracasei — 8.5±0.2 —
L. rhamnosus — — 7.8±0.2
E. coli 9.2±0.3 9.4±0.3 9.8±0.5**
B. breve 9.1±0.2 7.78±2.13 8.7±0.3*
B. longum 8.2±0.6 5.6±1.9*** 6.3±0.5***
S. aureus 7.4±0.3 6.3±0.3*** 6.6±0.5***
S. epidermidis 4.8±0.4 4.9±1.2 4.0±0.5**
C. perfringens 7.2±0.3 7.0±0.5 5.7±1.0***
Bacteroides 10.3±0.2 10.4±0.2 10.1±0.4
log10 CFU (colony-forming unit) given per gram of wet weight of feces. Data are presented as mean±s.d. Absence of specific bacterial strains in the gut microflora is
indicated by ‘‘—’’. The values for the HBF mice supplemented with probiotics were compared to HBF control mice, *, ** and *** indicate a significant difference at 95,
99 and 99.9% confidence levels, respectively.
Table II Short-chain fatty acid content in the cecum from the different groups
Amounts of SCFAs given in mmol per gram of dry Acetate Propionate Isobutyrate Butyrate Isovalerate
feces for each group
HBF (n=10) 77.6±17.6 22.3±4.3 0.9±0.2 3±0.6 2.1±0.6
HBF+L. paracasei (n=9) 52.3±23.6*** 22.2±10.8 1.2±0.5*** 1.5±0.8*** 2.7±1.2**
HBF+L. rhamnosus (n=9) 40.6±8*** 20.3±2.8 0.8±0.2 2.1±0.4*** 2.1±0.5
Data are presented in mmol per gram of dry feces and are presented as means±s.d. The amounts of SCFAs for the HBF mice supplemented with probiotics were
compared to HBF control mice, ** and *** indicate a significant difference at 99 and 99.9% confidence levels, respectively.
probiotic modulation. A statistically significant metabolic phosphorylcholine (GPC) and triglycerides in mice fed with
phenotype separation between untreated mice and probiotic both probiotics compared to controls (Figure 1B and E).
supplemented animals was observed as reflected by the Elevated choline levels were observed in plasma of mice fed
2
high value of QY for each model (Cloarec et al, 2005b; Table with L. rhamnosus and reduced plasma citrate levels were
III). The corresponding coefficients describing the most observed in mice fed with L. paracasei compared to controls.
important metabolites in plasma, liver, urine and fecal extracts
that contributed to group separation are also listed in
Supplementary Table 1. The area normalized intensities Fecal extract metabolic profiles
(101 a.u.) of representative metabolite signals are given as Marked changes were observed in the metabolic profiles of
means±s.d. in Table III. The O-PLS-DA coefficients plots are fecal extracts from all supplemented mice, for example relative
presented in Figure 1 using a back-scaling transformation and decreased concentrations of choline, acetate, ethanol, a range
projection to aid biomarker visualization (Cloarec et al, of putative N-acetylated metabolites (NAMs), unconjugated
2005b). The direction of the signals in the plots relative to bile acids (BAs) and tauro-conjugated bile acids (Figure 1C and
zero indicates positive or negative covariance with the F). Furthermore, relative higher levels of glucose, lysine and
probiotic-treated class. Each variable is plotted with a color polysaccharides were detected in the feces from mice fed with
code that indicates its discriminating power as calculated from probiotics. A relative increased level of n-caproate (chemical
the correlation matrix thus highlighting biomarker-rich shifts d at 0.89(t), 1.27(m), 1.63(q), 2.34(t)) appeared to be
spectral regions. associated with mice supplemented with L. paracasei.
Liver metabolic profiles
Livers of mice fed with L. paracasei showed relative decreases Urine metabolic profiles
in dimethylamine (DMA), trimethylamine (TMA), leucine, Urine samples of mice supplemented with both probiotics
isoleucine, glutamine, and glycogen and increased levels of showed relative increased concentrations of indoleacetylgly-
succinate and lactate (Figure 1A). Mice supplemented with cine (IAG), phenylacetylglycine (PAG), tryptamine and a
L. rhamnosus showed relative decreases in leucine and relative decrease in the levels of a-keto-isocaproate and citrate
isoleucine and relative increases in succinate, TMA and (Figure 1G and H). Relative increased concentrations of a
trimethylamine-N-oxide (TMAO) in the liver compared to mixture of putative glycolipids (UGLp, chemical shifts of
controls (Figure 1D). multiplets at d 0.89, 1.27, 1.56, 1.68, 2.15, 2.25, 3.10, 3.55,
3.60), N-acetyl-glycoproteins (NAGs) and a reduction in
3-hydroxy-isovalerate were also observed in mice supplemen-
Plasma metabolic profiles ted with L. paracasei compared to controls. Urine of mice fed
Plasma samples showed relative decreases in the levels of with L. rhamnosus showed a reduction in levels of creatine and
lipoproteins and increases in the concentrations of glycero- citrulline.
& 2008 EMBO and Nature Publishing Group Molecular Systems Biology 2008 3

4. Probiotics modulation of mammalian metabolism
F-PJ Martin et al
Table III Summary of influential metabolites for discriminating NMR spectra of liver, plasma, fecal extracts and urine
Metabolites Chemical shift and multiplicity HBF controls HBF+L. paracsei HBF+L. rhamnosus
2
Liver QY=21%, R2 =44%
X QY=41%, R2 =37%
2
X
Leu 0.92(t) 2.4±0.6 1.7±0.3*** 1.9±0.5*
Ileu 0.94 (t) 0.8±0.1 0.6±0.05*** 0.7±0.2
Lactate 1.32(d) 38.4±5.8 46.2±8.2a 39.2±9.7
Succinate 2.41(s) 0.2±0.1 1.0±0.6** 0.7±0.3*
MA 2.61(s) 0.1±0.06 0.04±0.002** 0.08±0.05
TMA 2.91 (s) 0.2±0.04 0.07±0.03*** 0.2±0.09
TMAO 3.27(s) 10.3±2.2 13.1±3.7 18.5±8.0**
Gln 2.44(m) 0.4±0.1 0.3±0.1* 0.3±0.1
Glycogen 5.38–5.45 3.4±1.9 1.5±0.6* 3.2±1.9
Plasma QY=44%, R2 =50%
2
X QY=51%, R2 =32%
2
X
Lipoproteins 0.84 (m) 13.7±1.8 10.1±0.9*** 9.8±4.2**
Citrate 2.65(d) 1.4±0.3 0.9±0.4** 1.1±0.2**
Choline 3.2(s) 11.6±2.6 16.2±5.7* 20.5±3.8***
GPC 3.22(s) 44.1±4.6 57.3±12.5** 68.1±11.2***
Glyceryls 3.91(m) 2.0±0.3 2.5±0.4** 2.7±0.4**
Feces QY=90%, R2 =48%
2
X QY=89%, R2 =49%
2
X
Caprylate 1.27(m) 2.5±0.1 3.5±0.2*** 2.4±0.1
Lys 3.00(m) 3.2±0.8 5.0±0.3*** 4.9±1.2**
Osides 5.42(m) 0.9±0.09 1.2±0.1*** 1.4±0.1***
Bile acids 0.72(s) 3.1±0.9 1.8±0.7** 2.0±0.6*
Ethanol 1.18(t) 2.5±0.1 2.0±0.08*** 1.9±0.09***
Choline 3.20(s) 48.0±19.5 11.3±4.1*** 20.3±10.9***
NAM 2.06(m) 7.1±1.0 5.4±0.3*** 5.5±0.3***
Acetate 1.91(s) 58.7±34.2 27.0±9.2** 32.9±12.9*
U1 3.71(s) 9.2±0.5 7.3±0.3*** 8.2±0.5***
Urine QY=91%, R2 =47%
2
X QY=59%, R2 =46%
2
X
IAG 7.55(d) 0.1±0.03 0.6±0.2*** 0.4±0.2**
PAG 7.37(m) 0.8±0.1 1.5±0.3*** 1.2±0.4*
Tryptamine 7.70(d) 0.1±0.04 0.4±0.1*** 0.2±0.1**
UGLp 1.27(m) 1.7±0.1 2.7±0.4*** 1.7±0.2
Glycero-metabolites 4.04 (m) 1.7±0.1 2.2±0.2*** 1.9±0.2*
NAG 2.04(s) 3.5±0.2 4.3±0.2*** 3.8±0.5
Butyrate 0.90(t) 6.9±0.8 5.2±0.7** 5.2±0.9**
a-keto-isocaproate 0.94(d) 13.8±4.6 6.1±2.3** 7.9±2.1**
Propionate 1.05(t) 0.9±0.2 0.8±0.04* 0.8±0.1
3-hydroxy-isovalerate 1.24(s) 3.0±0.4 2.1±0.4** 2.6±0.2
Citrate 2.55(d) 10.8±7.6 1.6±0.6** 2.2±0.9*
Creatine 3.92(s) 5.7±2.1 4.5±1.5 3.5±0.3**
Citrulline 1.88(m) 3.8±0.5 3.3±0.4 3.0±0.4**
O-PLS models were generated for comparing probiotics treated to HBF control mice using one predictive and two orthogonal components, RX2 value shows how
much of the variation is explained, QY2 value represents the predictability of the models, and relates to its statistical validity. Data are presented as area normalized
intensities (101 a.u.) of representative metabolite signals as means±s.d. The values for the HBF mice supplemented with probiotics were compared to HBF control
mice, a, *, ** and *** indicate a significant difference at 90, 95, 99 and 99.9% confidence levels, respectively.
s, singlet; d, doublet; t, triplet; q, quartet; m, multiplet; dd, doublet of doublets. For metabolite abbreviations, refer to the key in Figure 1, MA: methylamine.
UPLC-MS analysis of bile acids in ileal flushes Integration of multicompartment metabolic data
The proportion of bile acids in ileal flushes from the different using hierarchical-principal component analysis
groups are given in Table IV and are shown as mean±s.d. of A principal component analysis (PCA) model was initially
the percentage of the total bile acid content. O-PLS-DA of the constructed separately for the metabolic data from each
data set revealed that the relative concentrations of bile acids individual biological matrix (plasma, urine, liver, fecal extracts
obtained from unsupplemented HBF mice are separated from and bile acid composition in ileal flush; Figures 3 and 4). Three
those treated with probiotics, the correlation observed with principal components were calculated for each cross-validated
L. paracasei being more significant than with L. rhamnosus as PCA model, except for the plasma where four principal
2
noted by the values of the cross-validated model parameter QY components were retained to maximize the explained variance
(Figure 2). For example, HBF mice supplemented with R2X and the cross-validation parameter Q2 following the
L. paracasei showed strong correlations with higher amounts standard sevenfold cross-validation method (Cloarec et al,
of GCA, CDCA and UDCA and lower levels of a-MCA in the 2005b). These PCA models descriptors (R2X/Q2) were 0.70/
ileal flushes when compared to controls. HBF mice fed with 0.36 (plasma), 0.41/0.11 (urine), 0.80/0.71 (liver), 0.94/0.67
L. rhamnosus also showed higher levels of GCA associated (bile acid) and 0.61/0.42 (feces). The score vectors tb from
with lower levels of TUDCA and TCDCA in the ileal flushes each model were then assigned as new X-variables (Figure 3).
when compared to untreated HBF mice. Thus, the top level X-matrix contained 16 descriptors, denoted
4 Molecular Systems Biology 2008 & 2008 EMBO and Nature Publishing Group

5. Probiotics modulation of mammalian metabolism
F-PJ Martin et al
Figure 1 O-PLS-DA coefficient plots derived from 1H MAS NMR CPMG spectra of liver (A, D), 1H NMR CPMG spectra of plasma (B, E), 1H NMR standard spectra of
fecal extracts (C, F) and urine (G, H), indicating discrimination between HBF mice fed with probiotics (positive) and HBF control mice (negative). The color code
corresponds to the correlation coefficients of the variables with the classes. BAs, Bile acids; DMA, dimethylamine; Glc, glucose; Gln, glutamine; GPC,
glycerophosphorylcholine; IAG, indoleacetylglycine; Ileu, isoleucine; Leu, leucine; Lys, lysine; NAG, N-acetylated glycoproteins; NAM, N-acetylated metabolites;
Osides, glycosides; PAG, phenylacetylglycine; TBAs, taurine conjugated to bile acids; TMA, trimethylamine; TMAO, trimethylamine-N-oxide; UGLp, unidentified
glycolipids.
Pi (plasma PCs 1–4), Li (liver PCs 1–3), Ui (urine PCs 1–3), Bi (H-PCA; Westerhuis et al, 1998) model (R2X¼0.60) accounted
(bile acid PCs 1–3) and Fi (fecal PCs 1–3), which comprise only for 37 and 23% of the total variance in the combined multi
the systematic variation from each of the blocks/compart- compartment data respectively. The cross validation for the
ments. The two first principal components (p1 and p2) H-PCA model failed due to the high degree of orthogonality
calculated for the hierarchical-principal component analysis within the X-matrix, that is, within each of the blocks all variables
& 2008 EMBO and Nature Publishing Group Molecular Systems Biology 2008 5

6. Probiotics modulation of mammalian metabolism
F-PJ Martin et al
are orthogonal to each other, while each of the biological matrices samples have no weight in discriminating the bacterial
could be cross-validated at the individual level. supplementation from the corresponding controls in the global
The H-PCA scores plot illustrated a degree of clustering with model.
respect to the groups of HBF mice (Figure 4A). The To uncover variables contributing to the H-PCA super
corresponding H-PCA loadings plot indicated the contribution scores, the loadings at the base level PCA model were
of the 16 descriptors to the differences observed between the interrogated (Figure 4C–F). HBF mice supplemented with
samples in the H-PCA scores plot. Probiotic-supplemented probiotics showed higher concentrations of glucose, choline,
HBF animals are separated from the controls along the first GPC, glutamine, glutamate and lysine in the plasma profiles
principal component, and this arises from the main variations associated with elevated concentrations of glucose in the liver
modelled at the base level PCA of the individual plasma (P1, and higher levels of TUDCA and TCDCA in the ileal flushes.
P2), liver (L2, L3), ileal flush (B2) and urine (U1, U2) data sets Controls showed higher levels of lipoproteins in the plasma,
(Figure 4B). HBF mice fed with L. paracasei were separated elevated concentrations of lipids, glycogen, glutamine, gluta-
from those fed with L. rhamnosus along the second principal mate, alanine, TMAO and lactate in the liver, associated with
component, which was mainly due to the variations modelled higher levels of TCA and TbMCA in the ileal content.
at the base level PCA of plasma (P2, P3), liver (L1) and urine Unsupplemented HBF mice also showed elevated urinary
(U1, U3). Interestingly, the metabolic variations in the fecal excretions of creatine, citrate, citrulline, lysine, UGLp, NAG
and a-keto-isocaproate compared to animals fed with probio-
tics. Moreover, H-PCA also revealed that HBF mice treated
Table IV Bile acids composition in gut flushes for the different microbiota with L. rhamnosus had higher levels of hepatic lipids and
plasma lipoproteins but lower concentrations of lactate and
Microbiota/ HBF HBF+L. paracasei HBF+L. rhamnosus
amino acids in plasma and lower urinary excretion of PAG,
Bile acids
IAG, tryptamine and taurine than HBF mice treated with
CDCA ND 0.04±0.07 0.01±0.02 L. paracasei.
UDCA ND 0.1±0.1 ND
CA 0.4±0.5 0.3±0.7 0.3±0.2
oMCA ND 0.02±0.07 ND
aMCA 0.3±0.2 0.3±0.6 0.2±0.2
bMCA 0.9±0.7 1.3±2.7 0.8±0.7 Integration of correlations between bile acids and
GCA ND 0.1±0.1 0.1±.08 fecal flora
TCDCA 3.3±1 4±2.1 2.5±1.3
TUDC A 6.6±1.4 6.6±5.2 5.1±1.02 We further investigated the connections between fecal flora
TbMCA 50±4.8 49.2±7.5 50.5±5.2 and intestinal bile acids using a correlation analysis based
TCA 38±3.5 38±6.2 40.4±5 bipartite graphical modelling approach (see Materials and
methods; Figure 5) used previously to investigate the effects of
Relative composition in bile acids given in percentage of total bile acid content.
Species not detected with UPLC-MS experiment are shown as ND. The key is gut microbiome humanization in germfree mice (Martin et al,
given in UPLC-MS material and methods. 2007a). In the current study, we are working with a superior
0.4 0.9
HBF+L. paracasei
Class members hip
1 0.7
Covariance
r2
0.5
0.3
–1
HBF
–0.4 0.1
–3 0 5
CD
UD
CA
G
TC
TU
Tβ
TC
α-
β-
CA
Tcv 1
M
M
M
DC
DC
A
CA
CA
CA
CA
CA
A
A
0.8 0.3
HBF+L. rhamnosus
Class members hip
1
Covariance
r2
0.2
–1
HBF
–0.6 0.1
–2 0 2
CD
UD
CA
G
TC
TU
Tβ
TC
α-
β-
CA
M
M
M
DC
DC
A
Tcv 1
CA
CA
CA
CA
CA
A
A
Figure 2 O-PLS-DA coefficient plots derived from the bile acid composition obtained by UPLC-MS analysis of ileal flushes, which indicate discrimination between HBF
control mice (negative) and HBF mice treated with probiotics (positive), (A) L. paracasei and (B) L. rhamnosus. The color code corresponds to the correlation
coefficients of the variables. One predictive and one orthogonal component were calculated; the respective Q2 and R2 are (76.4, 52.2%) and (50.3, 51.2%).
Y X
6 Molecular Systems Biology 2008 & 2008 EMBO and Nature Publishing Group

7. Probiotics modulation of mammalian metabolism
F-PJ Martin et al
HIGH-LEVEL
PCA model score variables
tT
High level for investigating t and p
Integration of the data
pT
Collecting the scores
SUB-LEVEL
Bile
Urine t1 Plasma t2 Feces t3 Liver t4 t5 Separate PCA models for each block
acids
Sub level for investigating t and p
p1 p2 p3 p4 p5 Easy interpretation
Breaking down data
Bile Original raw data
Urine Plasma Feces Liver
acids
Logical Blocks
X1 X2 X3 X4 X5
Figure 3 Schematic overview of H-PCA modelling: (Gunnarsson et al, 2003). In the sublevel, each block of data XB is modelled locally by a PCA model. Each block is
summarized by one or more loading vectors pb and score vectors tp (‘super variables’), which can be combined to form a new data matrix than can then be modelled
using PCA, which generates the ‘super scores’ tT and the ‘super loadings’ pT. All conventional PCA statistics and diagnostics are retained.
H-PCA scores H-PCA loadings
5 TUDCA
0.9 L1
0.6
TCDCA
HBF
HBF + L. rhamnosus 0.3
B2
p2
t2
HBF-+L. paracasei 0 U3 βMCA
P2 αMCA CA
–0.0 CDCA
0.0 GCA
F B3
32
F1 UDCA
B1
F L2 P1
P4 B2 L3 TβMCA
U2 –0.3 TCA
–5 –0.3 P3 U1
–5 0 5 -0.1 -0.0 0.6 – 0.4 0.0 0.4
t1 p1 B1
0.03 0.04 0.05
Lipoproteins Lipids Creatine Citrate
Citrulline, Lys
0
0
P1
U1
L1
0
Choline, GPC
Glc Choline, GPC
–0.05 –0.03 Glc –0.05 UGLp
δ1H
8.7
8.2
7.7
7.2
6.7
6.2
5.7
5.2
3.9
3.4
2.9
2.4
1.9
1.4
0.9
3.9
3.4
2.9
2.4
1.9
1.4
0.9
δ1H δ1H
7.7
7.2
6.7
6.2
5.7
5.2
3.9
3.4
2.9
2.4
1.9
1.4
0.9
0.03 0.06 Glycogen 0.05
Glc Lipoproteins δ-keto-isocaproate
Glu , Gln
NAG Citrulline
0
L2
U2
P2
0
0
–0.05 Glu, Gln, Lys –0.05 –0.05
5.7
5.2
3.9
3.4
2.9
2.4
1.9
1.4
0.9
5.2
4.2
3.9
3.7
3.4
3.2
2.9
2.7
2.4
2.2
1.9
1.7
1.4
1.2
0.9
δ1H δ1H δ1H
8.2
7.7
7.2
6.7
6.2
3.9
3.4
2.9
2.4
1.9
1.4
0.9
0.12 0.07
Lactate TMAO Lactate 0.04
Taurine Citrate
Ala
Val , Leu , Ileu Glu , Gln
L3
P3
Glu , Gln Ala
U3
0
0
0
PAG, IAG, Trypt
–0.04
Pyruvate
–0.03 –0.04 Taurine
δ1H
5.2
4.2
3.9
3.7
3.4
3.2
2.9
2.7
2.4
2.2
1.9
1.7
1.4
1.2
0.9
5.2
3.9
3.4
2.9
2.4
1.9
1.4
0.9
δ1H δ1H
8.2
7.7
7.2
6.7
6.2
3.9
3.4
2.9
2.4
1.9
1.4
0.9
Figure 4 H-PCA scores (A) and loadings (B) plots for the two first components derived from scores of separate PCA constructed separately for the metabolic data
from each individual biological matrix from HBF mice (’), HBF-L. paracasei mice (E) and HBF-L. rhamnosus mice (J). These PCA models explained 94% (bile
acid, C), 70% (plasma, D), 80% (liver, E), 41% (urine, F) and 61% (feces, data not shown) of the total variation in the data, respectively. The systematic variation from
each of the block/compartment is summarized by its score vectors denoted Pi (plasma PCs 1–4), Li (liver PCs 1–3), Ui (urine PCs 1–3), Bi (bile acid PCs 1–3) and Fi (fecal
PCs 1–3), which can be combined to form a new data matrix than can then be modelled using PCA. The individual PCA loadings were color coded according to their
contribution to the H-PCA model in red (principal component 1) and in blue (principal component 2). The model has been calculated from Pareto scaled data using two
cross-validated PCs, R2X¼60%. Ala, alanine; see Figure 1.
model where all the major bacteria strains are identified, concentrations are interdependent such as in the case of
which was not possible when considering conventional substrate–product biochemical reactions. Additional pixel
microflora. Positive and negative correlations show the multi- maps of the correlation matrices are given to help interpreta-
colinearity between bile acids and gut bacteria, whose tion in Supplementary Figure 1.
& 2008 EMBO and Nature Publishing Group Molecular Systems Biology 2008 7

8. Probiotics modulation of mammalian metabolism
F-PJ Martin et al
A HBF
Ba E.c
0.6∗ TMCA –0.5
aMA
–0.5 C.p
B.l
–0.5
0.5
0.5 0.5
0.5 TCDCA
S.a 0.6∗ B.b
0.6∗
0.5 0.6∗ TCA
–0.5
CA 0.6∗
0.5
bMA 0.6∗
TUDCA
0.5
S.e 0.6∗
B HBF-L. paracasei UDCA C HBF-L. rhamnosus
–0.5
Ba E.c Ba E.c
–0.6∗
–0.5 0.7∗ La TMCA La
0.5 TMCA
aMA aMA
C.p C.p
B.l B.l
GCA GCA
0.5 –0.6∗
–0.5
CDCA
–0.7∗
0.5
0.5 TCDCA TCDCA
0.5 B.b 0.6∗
S.a –0.6∗ B.b S.a
0.5 TCA TCA
0.7∗∗∗
CA CA
–0.5
bMA 0.6∗ bMA
0.5
0.7∗ 0.5 TUDCA 0.6∗ TUDCA
0.5
S.e S.e
Figure 5 Integration of bile acid and fecal flora correlations. The bipartite graphs were derived from correlations between fecal flora and bile acids in each group: HBF
mice (A), HBF mice supplemented with L. paracasei (B) or L. rhamnosus (C). The cut-off value of 0.5 was applied to the absolute value of the coefficient |r| for
displaying the correlations between fecal flora and bile acids. Bile acids and fecal bacteria correspond to blue ellipse nodes and green rectangle nodes, respectively.
Edges are coded according to correlation value: positive and negative correlations are respectively displayed in blue and in red. aMA, a-muricholic acid; Ba,
Bacteroides; Bb, B. breve; Bl, B. longum; bMA, b-muricholic acid; CA, cholic acid; CDCA, chenodeoxycholic acid; Cp, C. perfringens; Ec, E. coli; GCA,
glycocholic acid; La, Lactobacillus probiotics; Sa, S. aureus; Se, S. epidermidis; TCA, taurocholic acid; TCDCA, taurochenodeoxycholic acid; TMCA, tauro-b-
muricholic acid; TUDCA, tauroursocholic acid; UDCA, ursocholic acid. * and *** indicate a statistically significant correlation at 95 and 99.9% confidence levels,
respectively.
Control HBF mice and HBF mice supplemented with Network analysis for HBF mice supplemented with
probiotics show remarkably different bile acid/fecal flora L. paracasei reveals that Lactobacilli supplementation resulted
correlation networks (Figure 5A–C), indicating that small in decreasing the functional links between Bifidobacteria and
modulations in the species composition of the microbiome can bile acids, while new significant correlations were observed
result in major functional ecological consequences. Network between bile acids, and Bacteroides, S. aureus, S. epidermidis
statistics reveal that microbiome/metabolome bipartite graphs and L. paracasei. Moreover, E. coli has several connections
from HBF mice supplemented with Lactobacilli show a totally with UDCA, aMCA and TCDCA. Interestingly, Bacteroides
different nodal structure (given for the cut-off value of 0.5). In shows functional correlations of opposite signs for TbMCA,
particular, we observed that in the network obtained from TUDCA and TCA when compared to Lactobacilli and
control HBF mice, the most connected bacteria are Bifidobac- Staphylococci.
teria, Staphylococci, and Clostridia, for which the variations When HBF mice received L. rhamnosus probiotic, the
are intrinsically correlated with the balance in tauro-conju- microbiome/metabolome network shows a significant lower
gated bile acids (TCDCA, TbMCA, TCA, TUDCA) and uncon- level of complexity (given for the cut-off value of 0.5). The
jugated bile acids (CA, aMCA, bMCA). In particular, the balance within B. breve, S. aureus and S. epidermidis appears
potentially harmful opportunist C. perfringens shows func- highly correlated to the composition in tauro-conjugated
tional correlation of opposite sign for TbMCA, TCDCA, aMCA bile acids (TbMCA, TCA) and unconjugated bile acids
and bMCA when compared to Bifidobacteria and S. aureus. (bMCA, CA).
8 Molecular Systems Biology 2008 & 2008 EMBO and Nature Publishing Group